Assessing the performance of computational predictors for estimating protein stability changes upon missense...

Area of Science:

• Computational biology
• Protein engineering
• Bioinformatics

Background:

• Missense mutations can alter protein stability, impacting protein engineering and disease understanding.
• Existing computational tools for predicting mutation effects on protein stability suffer from biases due to training and evaluation data.
• Identifying reliable features to discern mutation effects on protein stability is an ongoing challenge.

Purpose of the Study:

• To provide a comprehensive overview and benchmark of freely available computational tools for predicting protein stability changes upon missense mutations.
• To evaluate the performance of these tools considering mutation site, residue type, pH, and temperature variations.
• To offer guidance for developing improved next-generation bioinformatic tools.

Main Methods:

• Benchmarking of diverse, freely available predictive tools using three large, independent, mutation-level blind datasets (ThermoMutDB, iStable2.0, ProThermDB).
• Performance evaluation based on mutation site, mutant residue type, and varying pH and temperature conditions.
• Classification of mutations as stabilizing or destabilizing (∆∆G ≥ 0 or < 0).

Main Results:

• Predictor performance is significantly influenced by the mutation site and the type of mutant residue.
• Most predictors exhibit low performance at pH 6-8 and temperatures above 65°C, with iStable2.0 being an exception on the S630 dataset.
• Integrated predictors demonstrated superior performance compared to individual mechanistic or machine learning predictors across the evaluated datasets (S268, S630, S1342).

Conclusions:

• The study highlights the variability in computational tool performance based on mutation characteristics and environmental conditions.
• Integrated prediction approaches show promise for more accurate stability change predictions.
• Findings provide valuable insights for the design and development of more robust and reliable bioinformatic tools for predicting mutation effects on protein stability.